Computer Vision News - December 2022

36 GeoMedIA Best Paper brain organization with the precision we would like. In recent years, classical cortical surface registration algorithms have been extended to allow multiple cortical features that better reflect brain organization to drive registration to a common space. However, even these approaches are limited in their ability to fully align all subjects to a common space. Learning-based registration algorithms emerged as alternative frameworks on the grounds that they are better at learning the space of variation to tackle topographical variation and are more efficient in learning population-specific templates. In this work, we propose GeoMorph , the first deep- discrete learningframework that performs registration on the cortical surface . GeoMorph takes twocortical surfaces (individual subject and template) as inputs and aims to learn a smooth displacement field on the